Dynamic driving risk warning and risk avoidance method and system enabled by multi-dimensional risk field

Abstract

The present invention belongs to the field of road vehicle control and relates to a dynamic driving risk warning and avoidance method and system enabled by a multi-dimensional risk field. The method evaluates driving risks, issues warnings based on driving risks, and implements risk avoidance strategies. During driving risk assessment, driving environment information is acquired, a possible state domain is constructed, and the accident discrimination domain of a single accident is calculated based on the state vectors in the possible state domain, thereby determining the possible accident domain of the single accident and the severity of the single accident. Subsequently, the risk value of a single accident and the risk value of a single state domain at the current time step as well as the real-time risk value of a single accident and the real-time risk value of multiple accidents are calculated. When the real-time risk value exceeds the real-time risk threshold, a dynamic time series of the risk field is generated. This method can not only optimize the calibration complexity and accuracy of the risk field model, but also provide an effective risk avoidance strategy to ensure driving safety.

Description

Technical Field

[0001] The present invention belongs to the field of road vehicle control and relates to predicting or avoiding possible or imminent collisions, and specifically to a dynamic driving risk warning and avoidance method and system enabled by a multi-dimensional risk field. Background Art

[0002] With the rapid development of intelligent transportation systems, driving safety technology is gradually shifting from passive protection to active warning and control. Among the emerging approaches to assessing driving risks in recent years, those based on potential field theory (i.e., driving risk fields) can encompass multiple scenario elements and are applicable to complex scenarios, rather than being limited to specific scenarios like following a vehicle or changing lanes. Therefore, they offer comprehensive content and versatility. Furthermore, because risks are assessed across the entire road space near the vehicle, the resulting risk assessment can be directly used to formulate risk avoidance strategies and control plans.

[0003] However, existing driving risk field models simply borrow electric field theory from physics, describing the risk value as a risk assessment indicator that follows the inverse law of the distance square (or η power), and modifying it with many coefficients. This risk field model has the following problems:

[0004] First, the traditional driving risk field has many undetermined coefficients and their physical meanings are unclear. Among them, the frequently used undetermined coefficients include risk weight coefficients (such as the "risk field weight coefficient of the w-th obstacle vehicle" in the published Chinese patent application No. 202411927151.1 and the "speed difference weight factor" in the application No. 202411485296.0), risk attenuation coefficients (such as the "high-order coefficient" and "obstacle shape coefficient" in the published Chinese patent application No. 202411547580.6 and the "obstacle shape coefficient" in the application No. 2024114 85296.0, the "nonlinear factor" in application number 202211137773.5, and correction coefficients (e.g., the "speed-related undetermined coefficient" in published Chinese patent application number 202411927151.1, the "undetermined coefficient one" and "undetermined coefficient two" in application number 202410427400.4, the "road condition influencing factor" in application number 202411030279.8, and the "second undetermined coefficient" in application number 202411133691.2). These undetermined coefficients are highly coupled, making it difficult to design targeted experiments for them, and the calibration process can fall into the dilemma of multiple solutions. Furthermore, the physical meanings of the various undetermined coefficients in traditional driving risk fields are unclear and cannot be directly measured experimentally. Instead, they require repeated adjustments to match experimental results. Furthermore, the multiple coefficients in traditional driving risk field models are highly coupled, requiring simultaneous adjustment and repeated balancing, a cumbersome process. Furthermore, the risk value of traditional driving risk fields is generally based on power functions or exponential functions, which makes it have no clear physical meaning.

[0005] Second, traditional driving risk fields (those that follow the inverse distance squared or η-power laws or are based on a two-dimensional Gaussian distribution) can only assess collision risk, but not vehicle instability risk (e.g., published Chinese patents No. 202411030279.8, 202411242843.2, 202411133691.2, 202211137773.5, and 202411018302.1; published literature includes Zhan Ming's "Research on Lane Change Risk Decision-Making and Trajectory Prediction for Autonomous Vehicles Based on Inverse Reinforcement Learning," published by North China University of Technology in 2024; and Du Qian's "Research on Autonomous Driving Decision-Making Methods Based on Risk Assessment and Deep Reinforcement Learning," published by Qilu University of Technology in 2024). This model focuses solely on the spatial relationship between the ego vehicle and obstacles, but ignores the vehicle's dynamic state. However, vehicle instability (such as skidding) is strongly related to the vehicle's dynamic state. Even if a safe distance is maintained from obstacles on slippery roads, accidents caused by vehicle instability cannot be avoided. Chinese patent application No. 202110934200.4 discloses a method and device for comprehensive risk assessment of vehicle instability and collision under extreme working conditions. The proposed comprehensive analysis and evaluation method can evaluate the collision risk and instability risk in parallel by independently applying the traditional risk field and instability index at the same time. However, this method only evaluates the instability risk based on the current control information, resulting in a lack of foresight in risk prediction.

[0006] Third, in existing risk avoidance decision-making methods based on traditional risk fields, some low-risk locations are used as sampling points for the vehicle's risk avoidance trajectory (for example, Yang Chao et al. published "Risk Avoidance Decision Planning for Intelligent Driving Vehicles Based on Spatiotemporal Risks." Automotive Engineering, 2024, 46(06):975-984). However, this method does not consider the risks brought by other state vectors (speed, yaw angle, center of mass sideslip angle, etc.) in addition to position, which may lead to the misalignment of the optimization target of the risk avoidance strategy, and thus the risk avoidance strategy contains suboptimal or dangerous vehicle dynamic behaviors. Another risk avoidance decision-making method is to use the risk value as the optimization objective function of trajectory planning or the reward function of RL (reinforcement learning). However, without direct coupling to the control variables (acceleration, yaw rate, front wheel steering angle, etc.), the traditional risk field cannot characterize the risk propagation delay effect caused by the change of the control variables, nor does it quantify the difference in the regulatory effectiveness of the control variables on risk (for example, the effect of braking on longitudinal risk is significantly stronger than that of steering).

[0007] Therefore, it is necessary to optimize the calibration complexity and accuracy of the driving risk field model, and to proactively warn of risks based on the predicted risks and provide effective risk avoidance strategies to ensure driving safety. Summary of the Invention

[0008] In view of the shortcomings and deficiencies of the existing technology, the purpose of the present invention is to provide a dynamic driving risk warning and avoidance method enabled by a multi-dimensional risk field, so as to optimize the calibration complexity and accuracy of the risk field model, while being able to evaluate the skidding risk and provide an effective avoidance strategy to ensure driving safety.

[0009] To achieve the above object, the present invention adopts the following technical solutions:

[0010] A dynamic driving risk warning and risk avoidance method enabled by a multi-dimensional risk field, the method comprising the following steps:

[0011] Step 1. Assess driving risks:

[0012] Step 1.1. Obtain driving environment information from the vehicle perception system;

[0013] Step 1.2. Construct the possible state domain;

[0014] Calculate the state transition based on the information obtained in step 1.1 and the vehicle kinematic model and dynamic model; when calculating the state transition, stipulate that the control quantity of the vehicle kinematic model obeys a truncated mixed generalized Pareto distribution, obtain the probability distribution function of the discrete truncated mixed generalized Pareto distribution through discretization processing, and calculate the possible value of the state vector of the vehicle kinematic model at step (n+1) based on the possible value of the control quantity at step (n) that appears in the probability distribution of the discrete truncated mixed generalized Pareto distribution; then continue to calculate the possible value of the control quantity of the vehicle kinematic model at step (n+1), the possible value of the control quantity of the vehicle dynamics model at step (n), and the possible value of the state vector of the vehicle dynamics model at step (n+1);

[0015] The state vectors of all kinematics and dynamics calculated in step n+1 are recorded as the complete possible state domain of this step:

[0016] Calculate the cumulative probability of each state vector in the complete possible state domain in step n+1, filter possible states from the complete possible state domain based on the probability threshold, and construct the possible state domain; at the same time, normalize the cumulative probability to obtain the corrected probability under different state vectors;

[0017] Step 1.3. Calculate the accident discrimination domain of a single accident based on the state vectors in the possible state domain in step 1.2, thereby determining the possible accident domain of the single accident and the severity of the single accident;

[0018] Step 1.4. Calculate the risk value of a single accident and the risk value of a single state domain at the current time step based on the severity of the single accident at each time step and the modified probability of the vehicle under different state vectors. Also determine whether the time step upper limit has been reached. If so, proceed to step 1.5; if not, proceed to step 1.2.

[0019] Step 1.5. Calculate the real-time risk value for a single accident based on the risk value for a single accident at each time step. Calculate the real-time risk value for multiple accidents based on the risk value for a single state domain at each time step. Determine whether the real-time risk threshold is exceeded. If so, proceed to step 1.6. If not, terminate.

[0020] Step 1.6. Generate a real-time risk field, i.e., a dynamic time series of the risk field;

[0021] Step 2. Generate early warning information based on the real-time risk value of a single accident and the real-time risk value of multiple accidents;

[0022] Step 3. Control the vehicle to avoid danger:

[0023] Obtain the real-time risk value of a single accident and the real-time risk value of multiple accidents. If any real-time risk value exceeds the control threshold, the value is increased from n=1 to n=n. max -3 Obtain the state sampling point sequence that minimizes the real-time risk field at each step and the kinematic control sequence corresponding to the state transition step by step, generate a comprehensive control sequence based on the kinematic control sequence, and control the vehicle according to the comprehensive control sequence to achieve risk avoidance.

[0024] As a preferred embodiment of the present invention, the driving environment information in step 1.1 includes information about the vehicle and nearby vehicles, information about nearby obstacles, and information about nearby physical separations; the information about the vehicle and nearby vehicles includes the position X of the vehicle on the geodetic coordinate X axis, the position Y of the vehicle on the geodetic coordinate Y axis, the yaw angle ψ of the vehicle, the speed of the vehicle on the body coordinate X axis The acceleration of the vehicle on the X-axis of the body coordinate The vehicle's yaw rate The vehicle's front wheel steering angle δ f , the vehicle's center of mass side slip angle β; nearby obstacle information includes obstacle mass m O , obstacle length L O , obstacle width W O , the position of the obstacle on the X-axis of the geodetic coordinate X0, the position of the obstacle on the Y-axis of the geodetic coordinate Y0, and the placement angle of the obstacle ψ0; the nearby physical separation information includes the starting position X of the physical separation on the X-axis of the geodetic coordinate B , Physical separation of the starting position Y on the Y axis of the geodetic coordinate B 、Physical separation placement angle ψ B ;

[0025] The state vector s of the vehicle kinematic model in step 1.2 VK and control vector u VK They are: The control variable of the vehicle dynamics model is the front wheel steering angle δ f , state vector sVD for: The superscript T represents the transpose symbol.

[0026] As a preferred embodiment of the present invention, the accident judgment domain of a single accident in step 1.3 includes the accident judgment domain of sideslip, the accident judgment domain of collision with an obstacle, the accident judgment domain of collision with a physical partition, and the accident judgment domain of collision with other vehicles; the possible accident domain of a single accident is the intersection of the possible state domain and the accident judgment domain of a single accident, including the possible accident domain of sideslip, the possible accident domain of collision with an obstacle, the possible accident domain of collision with a physical partition, and the possible accident domain of collision with other vehicles; the severity of a single accident includes the severity of sideslip, the severity of collision with an obstacle, the severity of collision with a physical partition, and the severity of collision with other vehicles.

[0027] As a further preferred embodiment of the present invention, the state transition equation of the vehicle kinematic model is:

[0028]

[0029] Where subscript n is the time step index and τ is the time step length;

[0030] The state transition equation of the kinetic model is:

[0031]

[0032] in, is the state matrix, is the control matrix, is the front wheel equivalent cornering stiffness, is the equivalent cornering stiffness of the rear wheel, d f and d r are the tire model parameters of the front and rear wheels, a is the distance from the center of mass to the front axle, b is the distance from the center of mass to the rear axle, I z is the yaw moment of inertia.

[0033] As a preferred embodiment of the present invention, the accident discrimination domain of sideslip is:

[0034] Among them, β max is the critical value of the sideslip angle at the center of mass, is the critical value of yaw angular velocity, s VK,ego ,s VD,ego The state vector representing the kinematics and dynamics of the vehicle, β ego represents the sideslip angle of the vehicle's center of mass, represents the yaw rate of the vehicle;

[0035] The severity of a sideslip is: Where ω is the slip severity evaluation coefficient, represents the speed of the vehicle;

[0036] The accident judgment domain of collision with obstacles is:

[0037]

[0038] Where h is the projection axis, is the projection axis set used to determine the collision with obstacle j, Overlap is the interval overlap judgment function, Proj is the projection interval function, V ego is the vertex coordinate matrix of the vehicle, V Oj is the vertex coordinate matrix of obstacle j;

[0039] The severity of the collision with the obstacle is: Among them, m ego is the vehicle mass, m Oj is the mass of obstacle j;

[0040] The accident judgment domain for collision with physical separation is:

[0041]

[0042] in, V is the projection axis set used to determine the collision with the physical partition q. Bq is the vertex coordinate matrix of the physical partition q;

[0043] The severity of collisions with physical separation is:

[0044] The accident discrimination domain for collisions with other vehicles is:

[0045]

[0046] in, is the projection axis set used to determine the collision with vehicle i, V Vi is the vertex coordinate matrix of vehicle i, (s VK,i ,s VD,i ) is the state vector of vehicle i, is the possible state domain of vehicle i;

[0047] The severity of the collision with other vehicle i is:

[0048]

[0049] Among them, the longitudinal speed of vehicle i is and yaw angle ψ Vi Taken from the possible accident domain with vehicle i as the main body In state s ego The vehicle state s where the ego vehicle may collideVi , ψ ego is the yaw angle of the vehicle, m Vi represents the mass of vehicle i.

[0050] As a preferred embodiment of the present invention, the risk value of a single accident at step n is expressed as:

[0051]

[0052] Among them, r slip,n is the risk value of sideslip at step n, r Vi,n is the risk value of collision with vehicle i at step n, r Oj,n is the risk value of collision with obstacle j at step n, r Bq,n is the risk value of collision with physical partition q at step n, and At the nth step, the state vector is The severity of the sideslip and the severity of the collision with vehicle i, obstacle j, and physical partition q, Represents the state of the vehicle The corrected probability of The state of vehicle i is The corrected probability of represents the possible accident domain of sideslip at step n, represents the possible accident domain of collision with other vehicles i at step n, represents the possible accident domain of collision between vehicle i and the ego vehicle at step n, represents the possible accident domain of collision with obstacle j at step n, represents the possible accident domain of collision with physical partition q at step n;

[0053] The expression of the risk value of the single state domain at step n is:

[0054] As a preferred embodiment of the present invention, the expression of the real-time risk value of a single accident is:

[0055]

[0056] Where γ is the discount factor, r c,slip represents the real-time risk value of sideslip accident, r c,Vi represents the real-time risk value of collision with other vehicles i, r c,Oj represents the real-time risk value of collision with obstacle j, r c,Bq Represents the real-time risk value of collision accident with physical separation q; n max Represents the maximum time step;

[0057] The expression of multi-accident real-time risk value is:

[0058]

[0059] Among them, r c represents the real-time risk value of multiple accidents, r n Represents the single-state domain risk value at step n.

[0060] As a preferred embodiment of the present invention, the real-time risk field generated in step 1.6 is a 6-dimensional scalar field, which includes multiple accident risk values ​​obtained by taking several state vectors in the possible state domain of a certain time step as initial states and integrating the single state domain risk values ​​of all time steps after the time step, and is used to reflect the real-time risk situation of the vehicle in various possible states at the time step.

[0061] The present invention also provides a dynamic driving risk warning and avoidance system enabled by a multi-dimensional risk field, which is connected to and communicates with a vehicle perception system through a first external interface, is connected to and communicates with a digital instrument through a second external interface, and is connected to and communicates with a vehicle control system through a third external interface; the dynamic driving risk warning and avoidance system includes a driving risk assessment module, a warning information generation module, and a vehicle avoidance control module; the vehicle perception system is used to obtain vehicle perception information and forward it to the driving risk assessment module; the driving risk assessment module is used to assess driving risks and send risk assessment results to the warning information generation module and the vehicle avoidance control module; the driving risk During risk assessment, the real-time risk value of a single accident and the real-time risk value of multiple accidents are assessed to generate a dynamic time series of a multi-dimensional state risk field; the warning information generation module is used to obtain the real-time risk value of a single accident and the real-time risk value of multiple accidents from the driving risk assessment module, and generate corresponding warning information according to the real-time risk value of the single accident and the real-time risk value of multiple accidents; the vehicle risk avoidance control module is used to obtain the dynamic time series of the multi-dimensional state risk field from the driving risk assessment module and execute the vehicle risk avoidance control algorithm; the digital instrument is used to receive warning information from the warning information generation module and display the current risk; the vehicle control system is used to receive a comprehensive control sequence from the vehicle risk avoidance control module and control the vehicle according to the comprehensive control sequence.

[0062] As a preferred embodiment of the present invention, the driving risk assessment module includes a physical model, an event discrimination module, an event consequence assessment module, and a risk field calculation module; wherein the physical model includes a vehicle model, an obstacle model, and a physical separation model; the vehicle model is used to record the state of the vehicle and nearby vehicles and calculate the state transition of the vehicle, including the vehicle's kinematic model and the vehicle's dynamic model; the obstacle model is used to record nearby obstacles; and the physical separation model is used to record nearby physical separations;

[0063] The event determination module is used to determine the possibility of the ego vehicle colliding or skidding, and includes a collision with other vehicles determination module, a collision with obstacles determination module, a collision with physical separation determination module, and a skidding determination module; wherein the collision with other vehicles determination module is used to determine the possibility of the ego vehicle colliding with other vehicles; the collision with obstacles determination module is used to determine the possibility of the ego vehicle colliding with obstacles; the collision with physical separation determination module is used to determine the possibility of the ego vehicle colliding with physical separation; and the skidding determination module is used to determine the possibility of the ego vehicle skidding;

[0064] The event consequence assessment module is used to assess the consequences of a collision or sideslip of the ego vehicle, and includes a collision consequence assessment module for other vehicles, a collision consequence assessment module for obstacles, a collision consequence assessment module for physical separation, and a sideslip consequence assessment module; wherein the collision consequence assessment module for other vehicles is used to assess the consequences of a collision between the ego vehicle and other vehicles; the collision consequence assessment module for obstacles is used to assess the consequences of a collision between the ego vehicle and an obstacle; the collision consequence assessment module for physical separation is used to assess the consequences of a collision between the ego vehicle and a physical separation; and the sideslip consequence assessment module is used to assess the consequences of a sideslip of the ego vehicle;

[0065] The risk field calculation module is used to calculate the real-time risk value of a single accident and the real-time risk value of multiple accidents, and simultaneously generate a dynamic time series of a multi-dimensional state risk field.

[0066] Advantages and beneficial effects of the present invention:

[0067] (1) The present invention is based on the theory of vehicle kinematics and dynamics, and describes the risk evolution through the Markov process, which greatly reduces the number of undetermined coefficients and solves the problem of numerous undetermined coefficients and unclear physical meanings in the existing driving risk field model.

[0068] (2) The present invention's driving risk field model has only one undetermined coefficient: the slip severity evaluation proportional coefficient. This coefficient has a clear physical meaning, namely, the proportional coefficient of the speed change during the collision to the initial speed before the sideslip. This reduces the difficulty of constructing targeted experimental traffic scenarios when calibrating the coefficient. In addition, the physical meaning of the risk value of the present invention's multidimensional risk field is the expected value of the vehicle speed change before and after the collision. This makes the model not only more generalizable but also directly verifiable through experiments.

[0069] (3) The dynamic driving risk warning and avoidance method enabled by the multi-dimensional risk field in the present invention can not only quantitatively assess the collision risk, but also quantitatively assess the vehicle skidding risk. Therefore, it can more comprehensively describe various risks in complex coupled high-risk traffic scenarios. In addition, the assessment of instability risk in the present invention relies on the dynamic state of the future time step in the Markov process. Compared with the assessment of instability risk based on current control information, this method has better foresight.

[0070] (4) The multi-dimensional state risk field time series in the present invention contains the direct correspondence between the state and control quantity and the risk value at multiple moments in the future. When using this information to formulate a hedging strategy, it can not only significantly reduce the degree of misalignment of the optimization target, but also effectively deal with the risk propagation delay effect and the difference in risk adjustment efficiency. This method can provide reliable information support for the formulation of a hedging strategy. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] Figure 1 The architecture diagram of the dynamic driving risk warning and risk avoidance system that enables the multi-dimensional risk field of the present invention;

[0072] Figure 2 A flow chart of the dynamic driving risk warning and risk avoidance method that enables the multi-dimensional risk field of the present invention;

[0073] Figure 3 A flowchart for evaluating driving risks according to the present invention;

[0074] Figure 4 Construct a possible state domain flow chart for the present invention;

[0075] Figure 5 This is a flow chart of the present invention for identifying accidents and evaluating consequences. DETAILED DESCRIPTION

[0076] In order to enable those skilled in the art to better understand the technical solutions and advantages of the present invention, the present application is described in detail below with reference to the accompanying drawings, but this is not intended to limit the scope of protection of the present invention.

[0077] like Figure 1 As shown, this embodiment provides a dynamic driving risk warning and avoidance system enabled by a multi-dimensional risk field. The system is connected to and communicates with the vehicle perception system through a first external interface, is connected to and communicates with the digital instrument through a second external interface, and is connected to and communicates with the vehicle control system through a third external interface. The dynamic driving risk warning and avoidance system includes a driving risk assessment module, a warning information generation module, and a vehicle risk avoidance control module.

[0078] Among them, the vehicle perception system is used to obtain vehicle perception information and forward it to the driving risk assessment module; the vehicle perception information specifically includes: the type, size, position, angle, speed, acceleration, angular velocity of nearby vehicles, the size, position, angle of nearby obstacles, the starting point, angle, and length of nearby physical separations, the current position, angle, speed, acceleration, angular velocity, front wheel steering angle, sideslip angle of the center of mass, mass and yaw moment of inertia of the vehicle, and also includes the road adhesion coefficient.

[0079] The driving risk assessment module is used to obtain vehicle perception information from the vehicle perception system, assess driving risks, and send the risk assessment results to the warning information generation module and the vehicle risk avoidance control module; when assessing driving risk, it evaluates the real-time risk value of a single accident and the real-time risk value of multiple accidents to generate a dynamic time series of a multi-dimensional state risk field;

[0080] The warning information generation module is used to obtain the single accident real-time risk value and the multi-accident real-time risk value from the driving risk assessment module, and generate corresponding warning information based on the single accident real-time risk value and the multi-accident real-time risk value;

[0081] The vehicle risk avoidance control module is used to obtain the multi-dimensional state risk field dynamic time series from the driving risk assessment module and execute the vehicle risk avoidance control algorithm;

[0082] The digital instrument is used to receive warning information from the warning information generation module and display the current risk. The warning information specifically includes: collision warning information, sideslip warning information and high-risk road condition warning information;

[0083] The vehicle control system is used to receive a comprehensive control sequence from the vehicle risk avoidance control module and control the vehicle according to the comprehensive control sequence.

[0084] Furthermore, in this embodiment, the driving risk assessment module includes a physical model, an event discrimination module, an event consequence assessment module, and a risk field calculation module. The physical model includes a vehicle model, an obstacle model, and a physical separation model. The vehicle model is used to record the state of the vehicle and nearby vehicles and calculate vehicle state transitions. It includes a vehicle kinematic model and a vehicle dynamic model. Both the vehicle kinematic model and the vehicle dynamic model require the use of the same set of basic attribute parameters:

[0085] a V =(m,I z ,L OA ,W OA ) T ;

[0086] Among them, m is the vehicle mass, I z is the yaw moment of inertia, L OA is the total length of the vehicle, WOA is the total width of the vehicle, and the superscript T represents transposition;

[0087] The state vector s of the kinematic model VK and control vector u VK They are:

[0088]

[0089] Where X is the position of the vehicle on the X-axis of the geodetic coordinate system (the positive direction is the forward direction of the road), Y is the position of the vehicle on the Y-axis of the geodetic coordinate system (the positive direction is perpendicular to the road and points to the left in the right-handed coordinate system), and ψ is the yaw angle of the vehicle (the positive direction is the left turn direction). is the speed of the vehicle on the X-axis of the vehicle body coordinate (the positive direction is the direction of the vehicle head), is the acceleration of the vehicle on the X-axis of the body coordinate, is the vehicle's yaw rate;

[0090] The state transfer equation is:

[0091]

[0092] Here, subscript n is the time step index and τ is the time step length.

[0093] The control quantity of the dynamic model is the front wheel steering angle δ f , state vector s VD for: Where β is the sideslip angle of the center of mass;

[0094] The state transfer equation is:

[0095]

[0096]

[0097] in, is the state matrix, is the control matrix, is the front wheel equivalent cornering stiffness, is the equivalent cornering stiffness of the rear wheel, d f and d r are the tire model parameters of the front and rear wheels, a is the distance from the center of mass to the front axle, b is the distance from the center of mass to the rear axle, I z is the yaw moment of inertia.

[0098] The obstacle model is used to record nearby obstacles. The basic attribute parameters a of the obstacle model are O and the state vector s O They are:

[0099] aO =(m O ,L O ,W O ) T ;

[0100] s O =(X O ,Y O ,ψ O ) T ;

[0101] Among them, m o is the mass of the obstacle, L o is the obstacle length, W o is the obstacle width, X o is the position of the obstacle on the X-axis of the geodetic coordinates, o is the position of the obstacle on the Y axis of the geodetic coordinate, ψ O is the placement angle of the obstacle;

[0102] The physical separation model is used to record nearby physical separations. The basic attribute parameter of the physical separation model is only the physical separation length L. B , state vector s B for:

[0103] s B =(X B ,Y B ,ψ B ) T ;

[0104] Among them, X B The starting position of the physical separation on the X axis of the geodetic coordinate, B is the starting position of the physical separation on the Y axis of the geodetic coordinate, ψ B The placement angle for physical separation;

[0105] The event determination module is used to determine the possibility of the ego vehicle colliding or skidding, and includes a collision with other vehicles determination module, a collision with obstacles determination module, a collision with physical separation determination module, and a skidding determination module; wherein the collision with other vehicles determination module is used to determine the possibility of the ego vehicle colliding with other vehicles; the collision with obstacles determination module is used to determine the possibility of the ego vehicle colliding with obstacles; the collision with physical separation determination module is used to determine the possibility of the ego vehicle colliding with physical separation; and the skidding determination module is used to determine the possibility of the ego vehicle skidding;

[0106] The event consequence assessment module is used to assess the consequences of a collision or sideslip of the ego vehicle, and includes a collision consequence assessment module for other vehicles, a collision consequence assessment module for obstacles, a collision consequence assessment module for physical separation, and a sideslip consequence assessment module; wherein the collision consequence assessment module for other vehicles is used to assess the consequences of a collision between the ego vehicle and other vehicles; the collision consequence assessment module for obstacles is used to assess the consequences of a collision between the ego vehicle and an obstacle; the collision consequence assessment module for physical separation is used to assess the consequences of a collision between the ego vehicle and a physical separation; and the sideslip consequence assessment module is used to assess the consequences of a sideslip of the ego vehicle;

[0107] The risk field calculation module is used to calculate the real-time risk value of a single accident and the real-time risk value of multiple accidents, and simultaneously generate a dynamic time series of a multi-dimensional state risk field.

[0108] like Figures 2 to 4 As shown, the present invention also provides a dynamic driving risk warning and risk avoidance method enabled by a multi-dimensional risk field, the method comprising the following steps:

[0109] Step 1. Assess driving risks:

[0110] Step 1.1. Record risk source instances:

[0111] Risk source instances are risk source representation models formed by direct acquisition or algorithmic derivation based on real-time data acquired by the vehicle perception system. In this invention, risk sources are divided into three categories: nearby vehicles, nearby obstacles, and nearby physical separations.

[0112] In this embodiment, the method for recording risk source instances is: first, obtain driving environment information from the vehicle perception system, specifically including the type, size, position, angle, speed, acceleration and angular velocity of nearby vehicles, as well as the size, position and angle of nearby obstacles, and the starting point, angle and length of nearby physical separations.

[0113] Next, record the nearby vehicles and the basic attribute parameters a of the nearby vehicle i Vi,init , initial value of kinematic state vector s VKi,init and the initial value of the kinematic control variable u VKi,init ; Among the above parameters, the vehicle mass m in the basic attribute parameters Vi It cannot be directly obtained from the vehicle perception system. Its value is determined by the vehicle type, specifically:

[0114]

[0115] Record nearby obstacles and basic attribute parameters a of nearby obstacles j Oj and the state vector s Oj ; Among the above parameters, the mass m in the basic attribute parametersOj It cannot be directly obtained from the vehicle perception system. Its value is determined by the size of the obstacle, specifically:

[0116] m Oj =(1t / m 2 )L Oj W Oj ;

[0117] Among them, L Oj Obstacle O j Length, W Oj Obstacle O j width;

[0118] Record the nearby physical separation and the length L of the nearby physical separation q Bq and the state vector s Bq .

[0119] Step 1.2. Construct the possible state domain:

[0120] Possible state domain It is a set of vehicle states calculated by analyzing the current state of the vehicle or nearby vehicles through an algorithm, including the kinematic state vector s VK , also including the dynamic state vector s VD .

[0121] like Figure 4 As shown, in this embodiment, the specific steps of constructing the possible state domain are:

[0122] First determine whether n = 0. If so, proceed to step 1.2.1. Record the initial state of the vehicle. If not, go to step 1.2.2.

[0123] Specifically, the method for recording the initial state of the ego vehicle is as follows: first, obtain the motion sensor data, the current vehicle mass and the current yaw moment of inertia from the vehicle perception system, including the current position, angle, speed, acceleration, angular velocity, center of mass sideslip angle and front wheel steering angle of the ego vehicle. Next, record the basic attribute parameters a of the ego vehicle. V,ego,init , initial value of kinematic state vector s VK,ego,init , initial value of kinematic control variable u VK,ego,init , initial value of dynamic state vector s VD,ego,init and the initial value of the dynamic control variable u VD,ego,init ; Among the above parameters, the vehicle length L in the basic attribute parameters OA and vehicle width W OA The intrinsic dimensions of the vehicle are known, rather than acquired in real time;

[0124] Finally, the state vector and control value of the kinematic and dynamic models of the ego vehicle when n = 0 are set as follows:

[0125] s VK,ego,0 =s VK,ego,init

[0126] u V x ,ego,0 =u VK,ego,init

[0127] s VD,ego,0 =s VD,ego,init

[0128] u VD,ego,0 =u VD,ego,init

[0129] Step 1.2.2. Calculate state transition:

[0130] When constructing the possible state domain, the control variables of the vehicle kinematic model are stipulated to obey the truncated mixture generalized Pareto distribution (TMGPD). The probability density function of the acceleration is:

[0131]

[0132] The probability density function of the yaw rate is:

[0133]

[0134] in, is the acceleration of the vehicle on the X-axis of the body coordinate at step n, k 1,accel , σ 1,accel 、k 2,accel and σ 2,accel is the probability distribution parameter of acceleration, is the acceleration of the previous step (when n>0, take the estimated value ), is the yaw angular velocity of the vehicle at step n, k 1,angvel , σ 1,angvel 、k 2,angvel and σ 2,angvel is the probability distribution parameter of the yaw rate, is the yaw rate of the previous step (when n>0, take the estimated value ), f MGPD is the probability density function of the Mixture Generalized Pareto Distribution (MGPD), and N is the normalization constant. The probability distribution parameter k of the acceleration is 1,accel , σ 1,accel 、k 2,accel and σ 2,accelThe reference ranges of the values ​​are 0.85~0.95, 0.45~0.55, -0.05~-0.04 and 0.40~0.50; the probability distribution parameter k of the yaw angular velocity is 1,angvel , σ 1,angvel 、k 2,angvel and σ 2,angvel The reference ranges of the values ​​are 0.50~0.60, 2.40~2.80, 0.50~0.60, 2.40~2.80; f MGPD The formulas for and N are as follows:

[0135]

[0136] Where v is the observed value of the random variable, k1, k2, σ1 and σ2 are probability distribution parameters, and v min and v max are the lower and upper limits of the truncation interval respectively; in the above probability density function of acceleration In the above probability density function of yaw rate

[0137] Discretizing the above probability density distribution, the probability distribution function of the discrete truncated mixture generalized Pareto distribution (DTMGPD) is:

[0138]

[0139] Among them, ρ is the discrete interval index, v (ρ) is the ρth discrete value, ρ max is the number of discrete intervals, δ is the width of the discrete interval, ρ max , δ and v (ρ) The value of is:

[0140]

[0141] Among them, n max The upper limit of the time step is specified. T is the risk assessment time domain, τ is the time step length, and T and τ are set according to the performance of the on-board computer;

[0142] The method for calculating state transition is: first, for each possible value of the kinematic control quantity that appears in the discrete probability distribution Calculate the possible values ​​of the kinematic state vector at step n+1 according to the state transfer equation of the kinematic model

[0143] Next, the possible values ​​of the kinematic control variables for the n+1th step are calculated according to the Discrete Truncated Mixture Generalized Pareto Distribution (DTMGPD). Specifically, the ρth discrete value That is, the ρth possible value of the kinematic control quantity; a possible value combination of the kinematic control quantity in the n+1th step It is determined by the first 1,n+1 Possible values ​​of longitudinal acceleration and ρ 2,n+1 Possible values ​​of yaw rate composition.

[0144] Then reverse the possible values ​​of the front wheel steering angle in the nth step of the dynamic model:

[0145]

[0146] in, is the possible value of the center of mass sideslip angle at step n, represents the possible value of the yaw rate at step n, Represents the possible values ​​of the speed at step n, which have been obtained in the calculation state transfer step of step n (i.e. the previous step).

[0147] Finally, the possible value of the center of mass sideslip angle at step n+1 is calculated according to the state transfer equation of the dynamic model. Then the possible value of the dynamic state vector in step n+1 is And record all the kinematic and dynamic state vectors calculated in step n+1 as the complete possible state domain of this step:

[0148]

[0149] Among them, ρ 1,max,n+1 Represents the number of discrete intervals of longitudinal acceleration at step n+1, ρ 2,max,n+1 The number of discrete intervals representing the yaw rate at step n+1;

[0150] Note that when n = 0, all state vectors and control variables should be the observed values ​​described in step 1.2.1 rather than the estimated values.

[0151] Step 1.2.3. Calculate the cumulative probability:

[0152] Specifically, the method for calculating the cumulative probability is:

[0153]

[0154] in, At the n+1th step, the state vector is The cumulative probability of and is the probability that the kinematic control component takes a discrete value in the discrete distribution, At the nth step, the state vector is The cumulative probability of .

[0155] Step 1.2.4. Probabilistic pruning:

[0156] Specifically, the probability pruning method is as follows: first, possible states are filtered from the complete possible state domain based on the probability threshold:

[0157]

[0158] in, is the possible state domain of the n+1th step, ε is the probability threshold, K=20000 is the upper limit of the number of states, and the value of ε is:

[0159]

[0160] Next, we perform probability normalization, and the state vector at step n+1 is The corrected probability is:

[0161]

[0162] in, represent Any set of possible kinematic and dynamic states in Represents the state vector at step n+1 The cumulative probability of .

[0163] Step 1.3. Identify the incident and assess its consequences:

[0164] The accident identification method can be summarized as follows: Selecting a critical accident state subset from the "possible state domain"—specifically, the intersection of the "possible state domain" and the "accident identification domain"—is referred to as the "possible accident domain." The consequences of an accident are measured by accident severity, and this method uses the change in vehicle speed before and after the collision.

[0165] like Figure 5 As shown, the steps to identify an accident and assess its consequences are:

[0166] Step 1.3.1. Determine whether a sideslip has occurred. If so, proceed to Step 1.3.2. If not, determine whether there are any obstacles nearby. If so, proceed to Step 1.3.3. If no obstacles are present, determine whether there are any physical barriers nearby. If so, proceed to Step 1.3.5. If no physical barriers are present, determine whether there are any other vehicles nearby. If so, proceed to Step 1.3.7. If no other vehicles are present, terminate the process.

[0167] Specifically, the method for determining sideslip is as follows:

[0168] The accident judgment domain of sideslip is defined as:

[0169] Among them, β max is the critical value of the sideslip angle at the center of mass, is the critical value of yaw angular velocity. Note that the subscripts related to the time step and discrete interval are omitted in the above formula, s VK,ego ,s VD,ego The state vector representing the kinematics and dynamics of the vehicle, β ego represents the sideslip angle of the vehicle's center of mass, Represents the yaw rate of the vehicle.

[0170]

[0171] Where μ is the road adhesion coefficient, Represents the speed of the vehicle.

[0172] Finally, we get the possible accident domain of sideslip: in, is the possible state domain of the vehicle.

[0173] Step 1.3.2. Evaluate the consequences of a sideslip and end the process.

[0174] Specifically, the method for evaluating the consequences of sideslip is as follows:

[0175] The severity of a sideslip is defined as: Where ω is the proportional coefficient for slip severity evaluation.

[0176] Step 1.3.3. Determine collision with obstacles:

[0177] The method for determining collision with an obstacle is as follows:

[0178] First, the accident judgment domain of collision with obstacles is calculated as:

[0179]

[0180] Where h is the projection axis, is the projection axis set used to determine the collision with obstacle j, Overlap is the interval overlap judgment function, Proj is the projection interval function, V ego is the vertex coordinate matrix of the vehicle, V Oj is the vertex coordinate matrix of obstacle j.

[0181] The projection axis set used in the above equation to determine the collision with obstacle j is:

[0182]

[0183] in, and is the projection axis of the vehicle, and is the projection axis of obstacle j; the interval overlap discriminant function and the projection interval function are defined as:

[0184]

[0185] Wherein, the superscript T represents the transposition symbol, a1 is the lower limit of the first projection interval, b1 is the upper limit of the first projection interval, a2 is the lower limit of the second projection interval, and b2 represents the upper limit of the second projection interval;

[0186] The vertex coordinate matrix of the vehicle and the vertex coordinate matrix of the obstacle are:

[0187]

[0188] Finally, the possible accident domain of collision with obstacles is:

[0189] Step 1.3.4. Evaluate the consequences of a collision with an obstacle:

[0190] Specifically, the severity of the collision with the obstacle is: Among them, m ego is the vehicle mass, m 0j is the mass of obstacle j;

[0191] Step 1.3.5. Identify collisions with physical separation:

[0192] Specifically, the method for determining collision with physical separation is as follows:

[0193] First, the accident judgment domain of collision with physical separation is calculated as:

[0194]

[0195] in, V is the projection axis set used to determine the collision with the physical partition q. Bq is the vertex coordinate matrix of the physical partition q, and its specific value is:

[0196]

[0197] Finally, the possible accident domain for collision with the physical partition q is:

[0198] Step 1.3.6. Evaluate the consequences of collision with physical separation:

[0199] Specifically, the severity of collisions with physical separation is:

[0200] Step 1.3.7. Determine collision with other vehicles:

[0201] Specifically, the method for determining collision with other vehicles is as follows:

[0202] First, calculate the accident discrimination domain of collision with other vehicles:

[0203]

[0204] in, is the projection axis set used to determine the collision with vehicle i, V Vi is the vertex coordinate matrix of vehicle i, (s VK,i ,s VD,i ) is the state vector of vehicle i, is the possible state domain of vehicle i (calculated in the same way as step 1.2); the specific values ​​of the projection axis set and vertex coordinate matrix are:

[0205]

[0206] The possible accident domain of collision with other vehicles i is:

[0207] Finally, it is necessary to record the states of other vehicles when a collision may occur, and the states of vehicle i that may collide with the vehicle itself (s VK,i ,s VD,i ) constitutes the possible accident domain with vehicle i as the main body It also records the vehicle's Each state (s VK,i ,s VD,i )

[0208] Step 1.3.8. Evaluate the consequences of a collision with another vehicle:

[0209] Specifically, the severity of the collision with other vehicle i is:

[0210]

[0211] Among them, the longitudinal speed of vehicle i is and yaw angle ψ Vi Taken from In state s ego The vehicle state s where the ego vehicle may collide Vi , ψ ego is the yaw angle of the vehicle, m Vi represents the mass of vehicle i.

[0212] Step 1.4. Calculate the single-state domain risk value and determine whether the time step upper limit is reached. If so, go to step 1.5; if not, go to step 1.2;

[0213] The single-state domain risk value is the sum of the risk values ​​generated by all risk sources in all states within the state domain within a time step.

[0214] Specifically, the method for calculating the risk value of a single state domain is as follows: first, the risk value of a single accident at step n is calculated, and the expression is:

[0215]

[0216] Among them, r slip,n is the risk value of sideslip at step n, r Vi,n is the risk value of collision with vehicle i at step n, r Oj,n is the risk value of collision with obstacle j at step n, r Bq,n is the risk value of collision with physical partition q at step n, and At the nth step, the state vector is The severity of various accidents under different circumstances, Represents the state of the vehicle The corrected probability of The state of vehicle i is The corrected probability of represents the possible accident domain of sideslip at step n, represents the possible accident domain of collision with other vehicles i at step n, represents the possible accident domain of collision between vehicle i and the ego vehicle at step n, represents the possible accident domain of collision with obstacle j at step n, represents the possible accident domain of collision with physical partition q at step n, represents any combination of the vehicle's kinematic and dynamic states from various possible accident domains, Representatives taken from Any combination of kinematic and dynamic states of vehicle i in .

[0217] Next, calculate the risk value of the single-state domain at step n, and the expression is:

[0218]

[0219] Step 1.5. Calculate the real-time risk value and determine whether it exceeds the real-time risk threshold. If so, proceed to step 1.6; if not, end.

[0220] The real-time risk value of a single accident is the risk value obtained by integrating the risk values ​​of single accidents at all time steps, which can reflect the real-time risk of a single possible accident to the vehicle; the real-time risk value of multiple accidents is the multi-accident risk value obtained by integrating the risk values ​​of single state domains at all time steps, which can reflect the comprehensive real-time risk situation of the vehicle.

[0221] In this embodiment, the real-time risk value of a single accident is first calculated, and the expression is:

[0222]

[0223]

[0224] Where γ is the discount factor, r c,slip represents the real-time risk value of sideslip accident, r c,Vi represents the real-time risk value of collision with other vehicles i, r c,Oj represents the real-time risk value of collision with obstacle j, r c,Bq Represents the real-time risk value of a collision accident with the physical partition q.

[0225] Next, we calculate the real-time risk value of multiple accidents. The expression is:

[0226]

[0227] Among them, r c represents the real-time risk value of multiple accidents, r n Represents the single-state domain risk value at step n;

[0228] Step 1.6. Generate a real-time risk field, i.e., a dynamic time series of the risk field;

[0229] In this embodiment, the generated real-time risk field is a six-dimensional scalar field. It contains multiple accident risk values ​​derived from the initial state vectors in the possible state domain at a given time step, combined with the single-state domain risk values ​​from all time steps after that time step. This field reflects the real-time risk of the vehicle in various possible states at that time step. The dynamic time series of the risk field can reflect how the risk field changes over time.

[0230] Specifically, the method for generating the dynamic time series of the risk field is as follows: first, calculate the possible state domain at the nth step Medium state is the real-time risk value of multiple accidents in the initial state, and its expression is:

[0231]

[0232] Among them, r t is the single-state domain risk value at step t, Represents the possible state domain of the vehicle at step n.

[0233] Next, the real-time risk field at step n is recorded as:

[0234]

[0235] in, Indicates real-time risk field R n exist The value at .

[0236] In this embodiment, a dynamic time series of the risk field is generated:

[0237] Specifically, the risk field dynamic time series is recorded as: {R n},n∈{1,2,…,n max -3};

[0238] Among them, the upper limit of n is n max -3 is mainly considered when n is too large, the number of remaining steps n max -n is not sufficient to generate an accurate real-time risk field.

[0239] Step 2. Generate warning information based on the real-time risk value of a single accident and the real-time risk value of multiple accidents, including collision warning information, sideslip warning information, and high-risk road condition warning information;

[0240] In this embodiment, the method for generating collision warning information is as follows: first, a single accident real-time risk value and risk source instance of collision with obstacles, physical partitions and other vehicles are obtained from a driving risk assessment module.

[0241] Next, filter the single accident real-time risk value greater than or equal to the single accident real-time risk warning value r alert,single The risk source is generated and the warning message is generated: "[Obstacle / Physical partition / Vehicle] at [directly in front / left front / right front / directly behind / left rear / right rear] has a high collision risk, and the risk value is [r c,Oj / r c,Bq / r c,Vi ], please pay attention to avoid"; single accident real-time risk warning value r alert,single The reference range is 0.8 to 1.2, and the direction of the risk source is determined based on the geodetic coordinates of the vehicle and the risk source.

[0242] Finally, collision warning information is sent to the digital instrument panel to show the driver the current risk.

[0243] In this embodiment, the method for generating the sideslip warning information is as follows: first, a real-time risk value of a single side slip accident is obtained from a driving risk assessment module.

[0244] Next, determine whether the real-time risk value of a single accident due to sideslip is greater than or equal to the real-time risk warning value r alert,single , and generate warning information "Side slip risk is high, risk value is r c,slip , please hold the steering wheel firmly and slow down."

[0245] Finally, the sideslip warning information is sent to the digital instrument to show the driver the current risk.

[0246] In this embodiment, the method for generating high-risk road condition warning information is as follows: first, multiple accident real-time risk values ​​are obtained from a driving risk assessment module.

[0247] Next, determine whether the multi-accident real-time risk value is greater than or equal to the multi-accident real-time risk warning value r alert,multi , and generate warning information "entering a high-risk road area, the risk value is r c Please hold the steering wheel firmly, slow down and drive carefully to avoid accidents. alert,multi The reference range is 1.0 to 1.5.

[0248] Finally, high-risk road condition warning information is sent to the digital instrument to show the driver the current risk.

[0249] Step 3. Control the vehicle to avoid danger:

[0250] Specifically, in this embodiment, the method for controlling the vehicle to avoid danger is: first, obtaining a single accident real-time risk value and a multiple accident real-time risk value from a driving risk assessment module.

[0251] Next, determine whether the single accident real-time risk value is greater than or equal to the single accident real-time risk control threshold r ctrl,single , whether the multi-accident real-time risk value is greater than or equal to the multi-accident real-time risk control threshold r ctrl,multi ; Single accident real-time risk control threshold r ctrl,single The reference range is 2.0 to 3.0, and the multi-accident real-time risk control threshold r ctrl,multi The reference range is 3.0 to 4.0.

[0252] If any real-time risk value exceeds the control threshold, then the value increases from n=1 to n=n. max -3 Obtain the state sampling point sequence S and the kinematic control sequence U corresponding to the state transition that minimizes the real-time risk field of each step step by time VK .

[0253] For example, if the real-time risk field R at step n is n The minimum state sampling point is Then Record it into the state sampling point sequence S, and then make the vehicle go from state Transfer to state Kinematic control quantity Record into kinematic control sequence U VK middle.

[0254] Next, generate the comprehensive control sequence U integrated , the comprehensive control sequence consists of a series of comprehensive control quantities u ego,n The calculation method of the comprehensive control quantity is:

[0255]

[0256] Among them, β ego,n and is the sampling point in the state sequence The weight, and They are u in the kinematic control sequence VK,ego,n and u VK,ego,n+1 The weight.

[0257] In this embodiment, the vehicle is controlled according to a comprehensive control sequence to achieve risk avoidance.

[0258] The above description of the present invention using specific examples is intended only to facilitate understanding of the present invention and is not intended to limit the present invention. A person skilled in the art of the present invention may make several simple deductions, modifications, or substitutions based on the principles of the present invention. Therefore, the scope of protection of the present invention shall be subject to the scope of protection of the claims.

Claims

1. A dynamic driving risk warning and risk avoidance method enabled by a multi-dimensional risk field, characterized by: The method comprises the following steps: Step 1. Assess driving risks: Step 1.

1. Obtain driving environment information from the vehicle perception system; Step 1.

2. Construct the possible state domain; Calculate the state transition based on the information obtained in step 1.1 and the vehicle kinematic model and dynamic model; when calculating the state transition, stipulate that the control quantity of the vehicle kinematic model obeys a truncated mixed generalized Pareto distribution, obtain the probability distribution function of the discrete truncated mixed generalized Pareto distribution through discretization processing, and calculate the possible value of the state vector of the vehicle kinematic model at step (n+1) based on the possible value of the control quantity at step (n) that appears in the probability distribution of the discrete truncated mixed generalized Pareto distribution; then continue to calculate the possible value of the control quantity of the vehicle kinematic model at step (n+1), the possible value of the control quantity of the vehicle dynamics model at step (n), and the possible value of the state vector of the vehicle dynamics model at step (n+1); The state vectors of all kinematics and dynamics calculated in step n+1 are recorded as the complete possible state domain of this step: Calculate the cumulative probability of each state vector in the complete possible state domain in step n+1, filter possible states from the complete possible state domain based on the probability threshold, and construct the possible state domain; at the same time, normalize the cumulative probability to obtain the corrected probability under different state vectors; Step 1.

3. Calculate the accident discrimination domain of a single accident based on the state vectors in the possible state domain in step 1.2, thereby determining the possible accident domain of the single accident and the severity of the single accident; Step 1.

4. Calculate the risk value of a single accident and the risk value of a single state domain at the current time step based on the severity of the single accident at each time step and the modified probability of the vehicle under different state vectors. Also determine whether the time step upper limit has been reached. If so, proceed to step 1.5; if not, proceed to step 1.

2. Step 1.

5. Calculate the real-time risk value for a single accident based on the risk value for a single accident at each time step. Calculate the real-time risk value for multiple accidents based on the risk value for a single state domain at each time step. Determine whether the real-time risk threshold is exceeded. If so, proceed to step 1.

6. If not, terminate. Step 1.

6. Generate a real-time risk field, i.e., a dynamic time series of the risk field; Step 2. Generate early warning information based on the real-time risk value of a single accident and the real-time risk value of multiple accidents; Step 3. Control the vehicle to avoid danger: Obtain the real-time risk value of a single accident and the real-time risk value of multiple accidents. If any real-time risk value exceeds the control threshold, the value is increased from n=1 to n=n. max -3 Obtain the state sampling point sequence that minimizes the real-time risk field at each step and the kinematic control sequence corresponding to the state transition step by step, generate a comprehensive control sequence based on the kinematic control sequence, and control the vehicle according to the comprehensive control sequence to achieve risk avoidance.

2. The dynamic driving risk warning and risk avoidance method enabled by multi-dimensional risk field according to claim 1 is characterized in that: The driving environment information in step 1.1 includes the information of the vehicle and nearby vehicles, nearby obstacles, and nearby physical separations; the information of the vehicle and nearby vehicles includes the position X of the vehicle on the geodetic coordinate X axis, the position Y of the vehicle on the geodetic coordinate Y axis, the yaw angle ψ of the vehicle, and the speed of the vehicle on the body coordinate X axis. The acceleration of the vehicle on the X-axis of the body coordinate The vehicle's yaw rate The vehicle's front wheel steering angle δ f , the vehicle's center of mass side slip angle β; nearby obstacle information includes obstacle mass m O , obstacle length L O , obstacle width W O , the position of the obstacle on the X axis of the geodetic coordinate O , the position Y of the obstacle on the Y axis of the geodetic coordinate O , the obstacle placement angle ψ O ; Nearby physical separation information includes the starting position X of the physical separation on the X axis of the geodetic coordinate B , Physical separation of the starting position Y on the Y axis of the geodetic coordinate B 、Physical separation placement angle ψ B ; The state vector s of the vehicle kinematic model in step 1.2 VK and control vector u VK They are: The control variable of the vehicle dynamics model is the front wheel steering angle δ f , state vector s VD for: The superscript T represents the transposition symbol.

3. The dynamic driving risk warning and risk avoidance method enabled by multi-dimensional risk field according to claim 2 is characterized in that: The accident judgment domain of a single accident in step 1.3 includes the accident judgment domain of sideslip, the accident judgment domain of collision with an obstacle, the accident judgment domain of collision with a physical partition, and the accident judgment domain of collision with other vehicles. The possible accident domain of a single accident is the intersection of the possible state domain and the accident judgment domain of a single accident, including the possible accident domain of sideslip, the possible accident domain of collision with an obstacle, the possible accident domain of collision with a physical partition, and the possible accident domain of collision with other vehicles. The severity of a single accident includes the severity of sideslip, the severity of collision with an obstacle, the severity of collision with a physical partition, and the severity of collision with other vehicles.

4. The dynamic driving risk warning and risk avoidance method enabled by multi-dimensional risk field according to claim 3 is characterized in that: The state transition equation of the vehicle kinematic model is: Where subscript n is the time step index and τ is the time step length; The state transition equation of the kinetic model is: in, is the state matrix, is the control matrix, is the front wheel equivalent cornering stiffness, is the equivalent cornering stiffness of the rear wheel, d f and d r are the tire model parameters of the front and rear wheels, a is the distance from the center of mass to the front axle, b is the distance from the center of mass to the rear axle, I z is the yaw moment of inertia, and m is the vehicle mass.

5. The dynamic driving risk warning and risk avoidance method enabled by multi-dimensional risk field according to claim 4 is characterized in that: The accident judgment domain of sideslip is: Among them, β max is the critical value of the sideslip angle at the center of mass, is the critical value of yaw angular velocity, s VK,ego ,s VD,ego The state vector representing the kinematics and dynamics of the vehicle, β ego represents the sideslip angle of the vehicle's center of mass, represents the yaw rate of the vehicle; The severity of a sideslip is: Where ω is the slip severity evaluation coefficient, represents the speed of the vehicle; The accident judgment domain of collision with obstacles is: Where h is the projection axis, is the projection axis set used to determine the collision with obstacle j, Overlap is the interval overlap judgment function, Proj is the projection interval function, V ego is the vertex coordinate matrix of the vehicle, V Oj is the vertex coordinate matrix of obstacle j; The severity of the collision with the obstacle is: Among them, m ego is the vehicle mass, m Oj is the mass of obstacle j; The accident judgment domain for collision with physical separation is: in, V is the projection axis set used to determine the collision with the physical partition q. Bq is the vertex coordinate matrix of the physical partition q; The severity of collisions with physical separation is: The accident discrimination domain for collisions with other vehicles is: in, is the projection axis set used to determine the collision with vehicle i, V Vi is the vertex coordinate matrix of vehicle i, (s VK,i ,s VD,i ) is the state vector of vehicle i, is the possible state domain of vehicle i; The severity of the collision with other vehicle i is: Among them, the longitudinal speed of vehicle i is and yaw angle ψ Vi Taken from the possible accident domain with vehicle i as the main body In state s ego The vehicle state s where the ego vehicle may collide Vi , ψ ego is the yaw angle of the vehicle, m Vi represents the mass of vehicle i.

6. The dynamic driving risk warning and risk avoidance method enabled by multi-dimensional risk field according to claim 5 is characterized in that: The risk value of a single accident at step n is expressed as: Among them, r slip,n is the risk value of sideslip at step n, r Vi,n is the risk value of collision with vehicle i at step n, r Oj,n is the risk value of collision with obstacle j at step n, r Bq,n is the risk value of collision with physical partition q at step n, and At the nth step, the state vector is The severity of the sideslip and the severity of the collision with vehicle i, obstacle j, and physical partition q, Represents the state of the vehicle The corrected probability of The state of vehicle i is The corrected probability of represents the possible accident domain of sideslip at step n, represents the possible accident domain of collision with other vehicles i at step n, represents the possible accident domain of collision between vehicle i and the ego vehicle at step n, represents the possible accident domain of collision with obstacle j at step n, represents the possible accident domain of collision with physical partition q at step n; The expression of the risk value of the single state domain at step n is:

7. The dynamic driving risk warning and risk avoidance method enabled by multi-dimensional risk field according to claim 6 is characterized in that: The expression of the real-time risk value of a single accident is: Where γ is the discount factor, r c,slip represents the real-time risk value of sideslip accident, r c,Vi represents the real-time risk value of collision with other vehicles i, r c,Oj represents the real-time risk value of collision with obstacle j, r c,Bq Represents the real-time risk value of collision accident with physical separation q; n max Represents the maximum time step; The expression of multi-accident real-time risk value is: Among them, r c represents the real-time risk value of multiple accidents, r n Represents the single-state domain risk value at step n.

8. The dynamic driving risk warning and risk avoidance method enabled by multi-dimensional risk field according to claim 7 is characterized in that: The real-time risk field generated in step 1.6 is a 6-dimensional scalar field. It contains multiple accident risk values ​​derived from the initial state of several state vectors in the possible state domain of a certain time step and the single state domain risk values ​​of all time steps after that time step. It is used to reflect the real-time risk situation of the vehicle in various possible states at that time step.

9. A dynamic driving risk warning and avoidance system enabled by multi-dimensional risk fields, characterized by: The system is used to implement the dynamic driving risk warning and avoidance method enabled by the multi-dimensional risk field as described in any one of claims 1 to 8. The system is connected to and communicates with the vehicle perception system through a first external interface, is connected to and communicates with the digital instrument through a second external interface, and is connected to and communicates with the vehicle control system through a third external interface; the dynamic driving risk warning and avoidance system includes a driving risk assessment module, a warning information generation module, and a vehicle avoidance control module; the vehicle perception system is used to obtain vehicle perception information and forward it to the driving risk assessment module; the driving risk assessment module is used to assess driving risks and send risk assessment results to the warning information generation module and the vehicle avoidance control module; when assessing driving risks, the real-time risk value of a single accident and the real-time risk value of multiple accidents are assessed to generate a dynamic time series of a multi-dimensional state risk field; the warning information generation module is used to obtain the real-time risk value of a single accident and the real-time risk value of multiple accidents from the driving risk assessment module, and generate corresponding warning information according to the real-time risk value of a single accident and the real-time risk value of multiple accidents; The vehicle risk avoidance control module is used to obtain the dynamic time series of the multi-dimensional state risk field from the driving risk assessment module and execute the vehicle risk avoidance control algorithm; the digital instrument is used to receive warning information from the warning information generation module and display the current risk; the vehicle control system is used to receive the comprehensive control sequence from the vehicle risk avoidance control module and control the vehicle according to the comprehensive control sequence.

10. The dynamic driving risk warning and avoidance system enabled by multi-dimensional risk field according to claim 9 is characterized in that: The driving risk assessment module includes a physical model, an event discrimination module, an event consequence assessment module, and a risk field calculation module; wherein the physical model includes a vehicle model, an obstacle model, and a physical separation model; the vehicle model is used to record the state of the vehicle and nearby vehicles and calculate the state transition of the vehicle, including the vehicle's kinematic model and the vehicle's dynamic model; the obstacle model is used to record nearby obstacles; and the physical separation model is used to record nearby physical separations; The event determination module is used to determine the possibility of the ego vehicle colliding or skidding, and includes a collision with other vehicles determination module, a collision with obstacles determination module, a collision with physical separation determination module, and a skidding determination module; wherein the collision with other vehicles determination module is used to determine the possibility of the ego vehicle colliding with other vehicles; the collision with obstacles determination module is used to determine the possibility of the ego vehicle colliding with obstacles; the collision with physical separation determination module is used to determine the possibility of the ego vehicle colliding with physical separation; and the skidding determination module is used to determine the possibility of the ego vehicle skidding; The event consequence assessment module is used to assess the consequences of a collision or sideslip of the ego vehicle, and includes a collision consequence assessment module for other vehicles, a collision consequence assessment module for obstacles, a collision consequence assessment module for physical separation, and a sideslip consequence assessment module; wherein the collision consequence assessment module for other vehicles is used to assess the consequences of a collision between the ego vehicle and other vehicles; the collision consequence assessment module for obstacles is used to assess the consequences of a collision between the ego vehicle and an obstacle; the collision consequence assessment module for physical separation is used to assess the consequences of a collision between the ego vehicle and a physical separation; and the sideslip consequence assessment module is used to assess the consequences of a sideslip of the ego vehicle; The risk field calculation module is used to calculate the real-time risk value of a single accident and the real-time risk value of multiple accidents, and simultaneously generate a dynamic time series of a multi-dimensional state risk field.

Images